19d01e20adda0ab0f614fca763dfd1d479255535
   1/*
   2Format of STDIN stream:
   3
   4  stream ::= cmd*;
   5
   6  cmd ::= new_blob
   7        | new_commit
   8        | new_tag
   9        | reset_branch
  10        ;
  11
  12  new_blob ::= 'blob' lf
  13        mark?
  14    file_content;
  15  file_content ::= data;
  16
  17  new_commit ::= 'commit' sp ref_str lf
  18    mark?
  19    ('author' sp name '<' email '>' ts tz lf)?
  20    'committer' sp name '<' email '>' ts tz lf
  21    commit_msg
  22    ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
  23    ('merge' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)*
  24    file_change*
  25    lf;
  26  commit_msg ::= data;
  27
  28  file_change ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf
  29                | 'D' sp path_str lf
  30                ;
  31  mode ::= '644' | '755';
  32
  33  new_tag ::= 'tag' sp tag_str lf
  34    'from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf
  35        'tagger' sp name '<' email '>' ts tz lf
  36    tag_msg;
  37  tag_msg ::= data;
  38
  39  reset_branch ::= 'reset' sp ref_str lf
  40    ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
  41    lf;
  42
  43  checkpoint ::= 'checkpoint' lf
  44    lf;
  45
  46     # note: the first idnum in a stream should be 1 and subsequent
  47     # idnums should not have gaps between values as this will cause
  48     # the stream parser to reserve space for the gapped values.  An
  49         # idnum can be updated in the future to a new object by issuing
  50     # a new mark directive with the old idnum.
  51         #
  52  mark ::= 'mark' sp idnum lf;
  53
  54     # note: declen indicates the length of binary_data in bytes.
  55     # declen does not include the lf preceeding or trailing the
  56     # binary data.
  57     #
  58  data ::= 'data' sp declen lf
  59    binary_data
  60        lf;
  61
  62     # note: quoted strings are C-style quoting supporting \c for
  63     # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
  64         # is the signed byte value in octal.  Note that the only
  65     # characters which must actually be escaped to protect the
  66     # stream formatting is: \, " and LF.  Otherwise these values
  67         # are UTF8.
  68     #
  69  ref_str     ::= ref     | '"' quoted(ref)     '"' ;
  70  sha1exp_str ::= sha1exp | '"' quoted(sha1exp) '"' ;
  71  tag_str     ::= tag     | '"' quoted(tag)     '"' ;
  72  path_str    ::= path    | '"' quoted(path)    '"' ;
  73
  74  declen ::= # unsigned 32 bit value, ascii base10 notation;
  75  binary_data ::= # file content, not interpreted;
  76
  77  sp ::= # ASCII space character;
  78  lf ::= # ASCII newline (LF) character;
  79
  80     # note: a colon (':') must precede the numerical value assigned to
  81         # an idnum.  This is to distinguish it from a ref or tag name as
  82     # GIT does not permit ':' in ref or tag strings.
  83         #
  84  idnum   ::= ':' declen;
  85  path    ::= # GIT style file path, e.g. "a/b/c";
  86  ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
  87  tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
  88  sha1exp ::= # Any valid GIT SHA1 expression;
  89  hexsha1 ::= # SHA1 in hexadecimal format;
  90
  91     # note: name and email are UTF8 strings, however name must not
  92         # contain '<' or lf and email must not contain any of the
  93     # following: '<', '>', lf.
  94         #
  95  name  ::= # valid GIT author/committer name;
  96  email ::= # valid GIT author/committer email;
  97  ts    ::= # time since the epoch in seconds, ascii base10 notation;
  98  tz    ::= # GIT style timezone;
  99*/
 100
 101#include "builtin.h"
 102#include "cache.h"
 103#include "object.h"
 104#include "blob.h"
 105#include "tree.h"
 106#include "delta.h"
 107#include "pack.h"
 108#include "refs.h"
 109#include "csum-file.h"
 110#include "strbuf.h"
 111#include "quote.h"
 112
 113struct object_entry
 114{
 115        struct object_entry *next;
 116        unsigned long offset;
 117        unsigned type : TYPE_BITS;
 118        unsigned pack_id : 16;
 119        unsigned char sha1[20];
 120};
 121
 122struct object_entry_pool
 123{
 124        struct object_entry_pool *next_pool;
 125        struct object_entry *next_free;
 126        struct object_entry *end;
 127        struct object_entry entries[FLEX_ARRAY]; /* more */
 128};
 129
 130struct mark_set
 131{
 132        int shift;
 133        union {
 134                struct object_entry *marked[1024];
 135                struct mark_set *sets[1024];
 136        } data;
 137};
 138
 139struct last_object
 140{
 141        void *data;
 142        unsigned long len;
 143        unsigned long offset;
 144        unsigned int depth;
 145        unsigned no_free:1;
 146};
 147
 148struct mem_pool
 149{
 150        struct mem_pool *next_pool;
 151        char *next_free;
 152        char *end;
 153        char space[FLEX_ARRAY]; /* more */
 154};
 155
 156struct atom_str
 157{
 158        struct atom_str *next_atom;
 159        int str_len;
 160        char str_dat[FLEX_ARRAY]; /* more */
 161};
 162
 163struct tree_content;
 164struct tree_entry
 165{
 166        struct tree_content *tree;
 167        struct atom_str* name;
 168        struct tree_entry_ms
 169        {
 170                unsigned int mode;
 171                unsigned char sha1[20];
 172        } versions[2];
 173};
 174
 175struct tree_content
 176{
 177        unsigned int entry_capacity; /* must match avail_tree_content */
 178        unsigned int entry_count;
 179        unsigned int delta_depth;
 180        struct tree_entry *entries[FLEX_ARRAY]; /* more */
 181};
 182
 183struct avail_tree_content
 184{
 185        unsigned int entry_capacity; /* must match tree_content */
 186        struct avail_tree_content *next_avail;
 187};
 188
 189struct branch
 190{
 191        struct branch *table_next_branch;
 192        struct branch *active_next_branch;
 193        const char *name;
 194        unsigned long last_commit;
 195        struct tree_entry branch_tree;
 196        unsigned char sha1[20];
 197};
 198
 199struct tag
 200{
 201        struct tag *next_tag;
 202        const char *name;
 203        unsigned char sha1[20];
 204};
 205
 206struct dbuf
 207{
 208        void *buffer;
 209        size_t capacity;
 210};
 211
 212struct hash_list
 213{
 214        struct hash_list *next;
 215        unsigned char sha1[20];
 216};
 217
 218/* Stats and misc. counters */
 219static unsigned long max_depth = 10;
 220static unsigned long max_objects = -1;
 221static unsigned long max_packsize = -1;
 222static unsigned long alloc_count;
 223static unsigned long branch_count;
 224static unsigned long branch_load_count;
 225static unsigned long object_count;
 226static unsigned long marks_set_count;
 227static unsigned long object_count_by_type[1 << TYPE_BITS];
 228static unsigned long duplicate_count_by_type[1 << TYPE_BITS];
 229static unsigned long delta_count_by_type[1 << TYPE_BITS];
 230
 231/* Memory pools */
 232static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
 233static size_t total_allocd;
 234static struct mem_pool *mem_pool;
 235
 236/* Atom management */
 237static unsigned int atom_table_sz = 4451;
 238static unsigned int atom_cnt;
 239static struct atom_str **atom_table;
 240
 241/* The .pack file being generated */
 242static const char *base_name;
 243static unsigned int pack_id;
 244static char *idx_name;
 245static struct packed_git *pack_data;
 246static struct packed_git **all_packs;
 247static int pack_fd;
 248static unsigned long pack_size;
 249static unsigned char pack_sha1[20];
 250
 251/* Table of objects we've written. */
 252static unsigned int object_entry_alloc = 5000;
 253static struct object_entry_pool *blocks;
 254static struct object_entry *object_table[1 << 16];
 255static struct mark_set *marks;
 256static const char* mark_file;
 257
 258/* Our last blob */
 259static struct last_object last_blob;
 260
 261/* Tree management */
 262static unsigned int tree_entry_alloc = 1000;
 263static void *avail_tree_entry;
 264static unsigned int avail_tree_table_sz = 100;
 265static struct avail_tree_content **avail_tree_table;
 266static struct dbuf old_tree;
 267static struct dbuf new_tree;
 268
 269/* Branch data */
 270static unsigned long max_active_branches = 5;
 271static unsigned long cur_active_branches;
 272static unsigned long branch_table_sz = 1039;
 273static struct branch **branch_table;
 274static struct branch *active_branches;
 275
 276/* Tag data */
 277static struct tag *first_tag;
 278static struct tag *last_tag;
 279
 280/* Input stream parsing */
 281static struct strbuf command_buf;
 282static unsigned long next_mark;
 283static struct dbuf new_data;
 284static FILE* branch_log;
 285
 286
 287static void alloc_objects(unsigned int cnt)
 288{
 289        struct object_entry_pool *b;
 290
 291        b = xmalloc(sizeof(struct object_entry_pool)
 292                + cnt * sizeof(struct object_entry));
 293        b->next_pool = blocks;
 294        b->next_free = b->entries;
 295        b->end = b->entries + cnt;
 296        blocks = b;
 297        alloc_count += cnt;
 298}
 299
 300static struct object_entry* new_object(unsigned char *sha1)
 301{
 302        struct object_entry *e;
 303
 304        if (blocks->next_free == blocks->end)
 305                alloc_objects(object_entry_alloc);
 306
 307        e = blocks->next_free++;
 308        hashcpy(e->sha1, sha1);
 309        return e;
 310}
 311
 312static struct object_entry* find_object(unsigned char *sha1)
 313{
 314        unsigned int h = sha1[0] << 8 | sha1[1];
 315        struct object_entry *e;
 316        for (e = object_table[h]; e; e = e->next)
 317                if (!hashcmp(sha1, e->sha1))
 318                        return e;
 319        return NULL;
 320}
 321
 322static struct object_entry* insert_object(unsigned char *sha1)
 323{
 324        unsigned int h = sha1[0] << 8 | sha1[1];
 325        struct object_entry *e = object_table[h];
 326        struct object_entry *p = NULL;
 327
 328        while (e) {
 329                if (!hashcmp(sha1, e->sha1))
 330                        return e;
 331                p = e;
 332                e = e->next;
 333        }
 334
 335        e = new_object(sha1);
 336        e->next = NULL;
 337        e->offset = 0;
 338        if (p)
 339                p->next = e;
 340        else
 341                object_table[h] = e;
 342        return e;
 343}
 344
 345static unsigned int hc_str(const char *s, size_t len)
 346{
 347        unsigned int r = 0;
 348        while (len-- > 0)
 349                r = r * 31 + *s++;
 350        return r;
 351}
 352
 353static void* pool_alloc(size_t len)
 354{
 355        struct mem_pool *p;
 356        void *r;
 357
 358        for (p = mem_pool; p; p = p->next_pool)
 359                if ((p->end - p->next_free >= len))
 360                        break;
 361
 362        if (!p) {
 363                if (len >= (mem_pool_alloc/2)) {
 364                        total_allocd += len;
 365                        return xmalloc(len);
 366                }
 367                total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
 368                p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
 369                p->next_pool = mem_pool;
 370                p->next_free = p->space;
 371                p->end = p->next_free + mem_pool_alloc;
 372                mem_pool = p;
 373        }
 374
 375        r = p->next_free;
 376        /* round out to a pointer alignment */
 377        if (len & (sizeof(void*) - 1))
 378                len += sizeof(void*) - (len & (sizeof(void*) - 1));
 379        p->next_free += len;
 380        return r;
 381}
 382
 383static void* pool_calloc(size_t count, size_t size)
 384{
 385        size_t len = count * size;
 386        void *r = pool_alloc(len);
 387        memset(r, 0, len);
 388        return r;
 389}
 390
 391static char* pool_strdup(const char *s)
 392{
 393        char *r = pool_alloc(strlen(s) + 1);
 394        strcpy(r, s);
 395        return r;
 396}
 397
 398static void size_dbuf(struct dbuf *b, size_t maxlen)
 399{
 400        if (b->buffer) {
 401                if (b->capacity >= maxlen)
 402                        return;
 403                free(b->buffer);
 404        }
 405        b->capacity = ((maxlen / 1024) + 1) * 1024;
 406        b->buffer = xmalloc(b->capacity);
 407}
 408
 409static void insert_mark(unsigned long idnum, struct object_entry *oe)
 410{
 411        struct mark_set *s = marks;
 412        while ((idnum >> s->shift) >= 1024) {
 413                s = pool_calloc(1, sizeof(struct mark_set));
 414                s->shift = marks->shift + 10;
 415                s->data.sets[0] = marks;
 416                marks = s;
 417        }
 418        while (s->shift) {
 419                unsigned long i = idnum >> s->shift;
 420                idnum -= i << s->shift;
 421                if (!s->data.sets[i]) {
 422                        s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
 423                        s->data.sets[i]->shift = s->shift - 10;
 424                }
 425                s = s->data.sets[i];
 426        }
 427        if (!s->data.marked[idnum])
 428                marks_set_count++;
 429        s->data.marked[idnum] = oe;
 430}
 431
 432static struct object_entry* find_mark(unsigned long idnum)
 433{
 434        unsigned long orig_idnum = idnum;
 435        struct mark_set *s = marks;
 436        struct object_entry *oe = NULL;
 437        if ((idnum >> s->shift) < 1024) {
 438                while (s && s->shift) {
 439                        unsigned long i = idnum >> s->shift;
 440                        idnum -= i << s->shift;
 441                        s = s->data.sets[i];
 442                }
 443                if (s)
 444                        oe = s->data.marked[idnum];
 445        }
 446        if (!oe)
 447                die("mark :%lu not declared", orig_idnum);
 448        return oe;
 449}
 450
 451static struct atom_str* to_atom(const char *s, size_t len)
 452{
 453        unsigned int hc = hc_str(s, len) % atom_table_sz;
 454        struct atom_str *c;
 455
 456        for (c = atom_table[hc]; c; c = c->next_atom)
 457                if (c->str_len == len && !strncmp(s, c->str_dat, len))
 458                        return c;
 459
 460        c = pool_alloc(sizeof(struct atom_str) + len + 1);
 461        c->str_len = len;
 462        strncpy(c->str_dat, s, len);
 463        c->str_dat[len] = 0;
 464        c->next_atom = atom_table[hc];
 465        atom_table[hc] = c;
 466        atom_cnt++;
 467        return c;
 468}
 469
 470static struct branch* lookup_branch(const char *name)
 471{
 472        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 473        struct branch *b;
 474
 475        for (b = branch_table[hc]; b; b = b->table_next_branch)
 476                if (!strcmp(name, b->name))
 477                        return b;
 478        return NULL;
 479}
 480
 481static struct branch* new_branch(const char *name)
 482{
 483        unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 484        struct branch* b = lookup_branch(name);
 485
 486        if (b)
 487                die("Invalid attempt to create duplicate branch: %s", name);
 488        if (check_ref_format(name))
 489                die("Branch name doesn't conform to GIT standards: %s", name);
 490
 491        b = pool_calloc(1, sizeof(struct branch));
 492        b->name = pool_strdup(name);
 493        b->table_next_branch = branch_table[hc];
 494        b->branch_tree.versions[0].mode = S_IFDIR;
 495        b->branch_tree.versions[1].mode = S_IFDIR;
 496        branch_table[hc] = b;
 497        branch_count++;
 498        return b;
 499}
 500
 501static unsigned int hc_entries(unsigned int cnt)
 502{
 503        cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
 504        return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
 505}
 506
 507static struct tree_content* new_tree_content(unsigned int cnt)
 508{
 509        struct avail_tree_content *f, *l = NULL;
 510        struct tree_content *t;
 511        unsigned int hc = hc_entries(cnt);
 512
 513        for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
 514                if (f->entry_capacity >= cnt)
 515                        break;
 516
 517        if (f) {
 518                if (l)
 519                        l->next_avail = f->next_avail;
 520                else
 521                        avail_tree_table[hc] = f->next_avail;
 522        } else {
 523                cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 524                f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
 525                f->entry_capacity = cnt;
 526        }
 527
 528        t = (struct tree_content*)f;
 529        t->entry_count = 0;
 530        t->delta_depth = 0;
 531        return t;
 532}
 533
 534static void release_tree_entry(struct tree_entry *e);
 535static void release_tree_content(struct tree_content *t)
 536{
 537        struct avail_tree_content *f = (struct avail_tree_content*)t;
 538        unsigned int hc = hc_entries(f->entry_capacity);
 539        f->next_avail = avail_tree_table[hc];
 540        avail_tree_table[hc] = f;
 541}
 542
 543static void release_tree_content_recursive(struct tree_content *t)
 544{
 545        unsigned int i;
 546        for (i = 0; i < t->entry_count; i++)
 547                release_tree_entry(t->entries[i]);
 548        release_tree_content(t);
 549}
 550
 551static struct tree_content* grow_tree_content(
 552        struct tree_content *t,
 553        int amt)
 554{
 555        struct tree_content *r = new_tree_content(t->entry_count + amt);
 556        r->entry_count = t->entry_count;
 557        r->delta_depth = t->delta_depth;
 558        memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
 559        release_tree_content(t);
 560        return r;
 561}
 562
 563static struct tree_entry* new_tree_entry()
 564{
 565        struct tree_entry *e;
 566
 567        if (!avail_tree_entry) {
 568                unsigned int n = tree_entry_alloc;
 569                total_allocd += n * sizeof(struct tree_entry);
 570                avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
 571                while (n-- > 1) {
 572                        *((void**)e) = e + 1;
 573                        e++;
 574                }
 575                *((void**)e) = NULL;
 576        }
 577
 578        e = avail_tree_entry;
 579        avail_tree_entry = *((void**)e);
 580        return e;
 581}
 582
 583static void release_tree_entry(struct tree_entry *e)
 584{
 585        if (e->tree)
 586                release_tree_content_recursive(e->tree);
 587        *((void**)e) = avail_tree_entry;
 588        avail_tree_entry = e;
 589}
 590
 591static void yread(int fd, void *buffer, size_t length)
 592{
 593        ssize_t ret = 0;
 594        while (ret < length) {
 595                ssize_t size = xread(fd, (char *) buffer + ret, length - ret);
 596                if (!size)
 597                        die("Read from descriptor %i: end of stream", fd);
 598                if (size < 0)
 599                        die("Read from descriptor %i: %s", fd, strerror(errno));
 600                ret += size;
 601        }
 602}
 603
 604static void start_packfile()
 605{
 606        struct packed_git *p;
 607        struct pack_header hdr;
 608
 609        idx_name = xmalloc(strlen(base_name) + 11);
 610        p = xcalloc(1, sizeof(*p) + strlen(base_name) + 13);
 611        sprintf(p->pack_name, "%s%5.5i.pack", base_name, pack_id + 1);
 612        sprintf(idx_name, "%s%5.5i.idx", base_name, pack_id + 1);
 613
 614        pack_fd = open(p->pack_name, O_RDWR|O_CREAT|O_EXCL, 0666);
 615        if (pack_fd < 0)
 616                die("Can't create %s: %s", p->pack_name, strerror(errno));
 617        p->pack_fd = pack_fd;
 618
 619        hdr.hdr_signature = htonl(PACK_SIGNATURE);
 620        hdr.hdr_version = htonl(2);
 621        hdr.hdr_entries = 0;
 622        write_or_die(pack_fd, &hdr, sizeof(hdr));
 623
 624        pack_data = p;
 625        pack_size = sizeof(hdr);
 626        object_count = 0;
 627
 628        all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
 629        all_packs[pack_id] = p;
 630}
 631
 632static void fixup_header_footer()
 633{
 634        SHA_CTX c;
 635        char hdr[8];
 636        unsigned long cnt;
 637        char *buf;
 638
 639        if (lseek(pack_fd, 0, SEEK_SET) != 0)
 640                die("Failed seeking to start: %s", strerror(errno));
 641
 642        SHA1_Init(&c);
 643        yread(pack_fd, hdr, 8);
 644        SHA1_Update(&c, hdr, 8);
 645
 646        cnt = htonl(object_count);
 647        SHA1_Update(&c, &cnt, 4);
 648        write_or_die(pack_fd, &cnt, 4);
 649
 650        buf = xmalloc(128 * 1024);
 651        for (;;) {
 652                size_t n = xread(pack_fd, buf, 128 * 1024);
 653                if (n <= 0)
 654                        break;
 655                SHA1_Update(&c, buf, n);
 656        }
 657        free(buf);
 658
 659        SHA1_Final(pack_sha1, &c);
 660        write_or_die(pack_fd, pack_sha1, sizeof(pack_sha1));
 661}
 662
 663static int oecmp (const void *a_, const void *b_)
 664{
 665        struct object_entry *a = *((struct object_entry**)a_);
 666        struct object_entry *b = *((struct object_entry**)b_);
 667        return hashcmp(a->sha1, b->sha1);
 668}
 669
 670static void write_index(const char *idx_name)
 671{
 672        struct sha1file *f;
 673        struct object_entry **idx, **c, **last, *e;
 674        struct object_entry_pool *o;
 675        unsigned int array[256];
 676        int i;
 677
 678        /* Build the sorted table of object IDs. */
 679        idx = xmalloc(object_count * sizeof(struct object_entry*));
 680        c = idx;
 681        for (o = blocks; o; o = o->next_pool)
 682                for (e = o->next_free; e-- != o->entries;)
 683                        if (pack_id == e->pack_id)
 684                                *c++ = e;
 685        last = idx + object_count;
 686        if (c != last)
 687                die("internal consistency error creating the index");
 688        qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
 689
 690        /* Generate the fan-out array. */
 691        c = idx;
 692        for (i = 0; i < 256; i++) {
 693                struct object_entry **next = c;;
 694                while (next < last) {
 695                        if ((*next)->sha1[0] != i)
 696                                break;
 697                        next++;
 698                }
 699                array[i] = htonl(next - idx);
 700                c = next;
 701        }
 702
 703        f = sha1create("%s", idx_name);
 704        sha1write(f, array, 256 * sizeof(int));
 705        for (c = idx; c != last; c++) {
 706                unsigned int offset = htonl((*c)->offset);
 707                sha1write(f, &offset, 4);
 708                sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
 709        }
 710        sha1write(f, pack_sha1, sizeof(pack_sha1));
 711        sha1close(f, NULL, 1);
 712        free(idx);
 713}
 714
 715static void end_packfile()
 716{
 717        struct packed_git *old_p = pack_data, *new_p;
 718
 719        if (object_count) {
 720                fixup_header_footer();
 721                write_index(idx_name);
 722                fprintf(stdout, "%s\n", old_p->pack_name);
 723                fflush(stdout);
 724
 725                /* Register the packfile with core git's machinary. */
 726                new_p = add_packed_git(idx_name, strlen(idx_name), 1);
 727                if (!new_p)
 728                        die("core git rejected index %s", idx_name);
 729                new_p->windows = old_p->windows;
 730                new_p->pack_fd = old_p->pack_fd;
 731                all_packs[pack_id++] = new_p;
 732                install_packed_git(new_p);
 733        }
 734        else {
 735                close(pack_fd);
 736                unlink(old_p->pack_name);
 737        }
 738        free(old_p);
 739        free(idx_name);
 740
 741        /* We can't carry a delta across packfiles. */
 742        free(last_blob.data);
 743        last_blob.data = NULL;
 744        last_blob.len = 0;
 745        last_blob.offset = 0;
 746        last_blob.depth = 0;
 747}
 748
 749static void checkpoint()
 750{
 751        end_packfile();
 752        start_packfile();
 753}
 754
 755static size_t encode_header(
 756        enum object_type type,
 757        size_t size,
 758        unsigned char *hdr)
 759{
 760        int n = 1;
 761        unsigned char c;
 762
 763        if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
 764                die("bad type %d", type);
 765
 766        c = (type << 4) | (size & 15);
 767        size >>= 4;
 768        while (size) {
 769                *hdr++ = c | 0x80;
 770                c = size & 0x7f;
 771                size >>= 7;
 772                n++;
 773        }
 774        *hdr = c;
 775        return n;
 776}
 777
 778static int store_object(
 779        enum object_type type,
 780        void *dat,
 781        size_t datlen,
 782        struct last_object *last,
 783        unsigned char *sha1out,
 784        unsigned long mark)
 785{
 786        void *out, *delta;
 787        struct object_entry *e;
 788        unsigned char hdr[96];
 789        unsigned char sha1[20];
 790        unsigned long hdrlen, deltalen;
 791        SHA_CTX c;
 792        z_stream s;
 793
 794        hdrlen = sprintf((char*)hdr,"%s %lu",type_names[type],datlen) + 1;
 795        SHA1_Init(&c);
 796        SHA1_Update(&c, hdr, hdrlen);
 797        SHA1_Update(&c, dat, datlen);
 798        SHA1_Final(sha1, &c);
 799        if (sha1out)
 800                hashcpy(sha1out, sha1);
 801
 802        e = insert_object(sha1);
 803        if (mark)
 804                insert_mark(mark, e);
 805        if (e->offset) {
 806                duplicate_count_by_type[type]++;
 807                return 1;
 808        }
 809
 810        if (last && last->data && last->depth < max_depth) {
 811                delta = diff_delta(last->data, last->len,
 812                        dat, datlen,
 813                        &deltalen, 0);
 814                if (delta && deltalen >= datlen) {
 815                        free(delta);
 816                        delta = NULL;
 817                }
 818        } else
 819                delta = NULL;
 820
 821        memset(&s, 0, sizeof(s));
 822        deflateInit(&s, zlib_compression_level);
 823        if (delta) {
 824                s.next_in = delta;
 825                s.avail_in = deltalen;
 826        } else {
 827                s.next_in = dat;
 828                s.avail_in = datlen;
 829        }
 830        s.avail_out = deflateBound(&s, s.avail_in);
 831        s.next_out = out = xmalloc(s.avail_out);
 832        while (deflate(&s, Z_FINISH) == Z_OK)
 833                /* nothing */;
 834        deflateEnd(&s);
 835
 836        /* Determine if we should auto-checkpoint. */
 837        if ((object_count + 1) > max_objects
 838                || (object_count + 1) < object_count
 839                || (pack_size + 60 + s.total_out) > max_packsize
 840                || (pack_size + 60 + s.total_out) < pack_size) {
 841
 842                /* This new object needs to *not* have the current pack_id. */
 843                e->pack_id = pack_id + 1;
 844                checkpoint();
 845
 846                /* We cannot carry a delta into the new pack. */
 847                if (delta) {
 848                        free(delta);
 849                        delta = NULL;
 850                }
 851                memset(&s, 0, sizeof(s));
 852                deflateInit(&s, zlib_compression_level);
 853                s.next_in = dat;
 854                s.avail_in = datlen;
 855                s.avail_out = deflateBound(&s, s.avail_in);
 856                s.next_out = out;
 857                while (deflate(&s, Z_FINISH) == Z_OK)
 858                        /* nothing */;
 859                deflateEnd(&s);
 860        }
 861
 862        e->type = type;
 863        e->pack_id = pack_id;
 864        e->offset = pack_size;
 865        object_count++;
 866        object_count_by_type[type]++;
 867
 868        if (delta) {
 869                unsigned long ofs = e->offset - last->offset;
 870                unsigned pos = sizeof(hdr) - 1;
 871
 872                delta_count_by_type[type]++;
 873                last->depth++;
 874
 875                hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
 876                write_or_die(pack_fd, hdr, hdrlen);
 877                pack_size += hdrlen;
 878
 879                hdr[pos] = ofs & 127;
 880                while (ofs >>= 7)
 881                        hdr[--pos] = 128 | (--ofs & 127);
 882                write_or_die(pack_fd, hdr + pos, sizeof(hdr) - pos);
 883                pack_size += sizeof(hdr) - pos;
 884        } else {
 885                if (last)
 886                        last->depth = 0;
 887                hdrlen = encode_header(type, datlen, hdr);
 888                write_or_die(pack_fd, hdr, hdrlen);
 889                pack_size += hdrlen;
 890        }
 891
 892        write_or_die(pack_fd, out, s.total_out);
 893        pack_size += s.total_out;
 894
 895        free(out);
 896        if (delta)
 897                free(delta);
 898        if (last) {
 899                if (last->data && !last->no_free)
 900                        free(last->data);
 901                last->data = dat;
 902                last->offset = e->offset;
 903                last->len = datlen;
 904        }
 905        return 0;
 906}
 907
 908static void *gfi_unpack_entry(
 909        struct object_entry *oe,
 910        unsigned long *sizep)
 911{
 912        static char type[20];
 913        struct packed_git *p = all_packs[oe->pack_id];
 914        if (p == pack_data)
 915                p->pack_size = pack_size + 20;
 916        return unpack_entry(p, oe->offset, type, sizep);
 917}
 918
 919static const char *get_mode(const char *str, unsigned int *modep)
 920{
 921        unsigned char c;
 922        unsigned int mode = 0;
 923
 924        while ((c = *str++) != ' ') {
 925                if (c < '0' || c > '7')
 926                        return NULL;
 927                mode = (mode << 3) + (c - '0');
 928        }
 929        *modep = mode;
 930        return str;
 931}
 932
 933static void load_tree(struct tree_entry *root)
 934{
 935        unsigned char* sha1 = root->versions[1].sha1;
 936        struct object_entry *myoe;
 937        struct tree_content *t;
 938        unsigned long size;
 939        char *buf;
 940        const char *c;
 941
 942        root->tree = t = new_tree_content(8);
 943        if (is_null_sha1(sha1))
 944                return;
 945
 946        myoe = find_object(sha1);
 947        if (myoe) {
 948                if (myoe->type != OBJ_TREE)
 949                        die("Not a tree: %s", sha1_to_hex(sha1));
 950                t->delta_depth = 0;
 951                buf = gfi_unpack_entry(myoe, &size);
 952        } else {
 953                char type[20];
 954                buf = read_sha1_file(sha1, type, &size);
 955                if (!buf || strcmp(type, tree_type))
 956                        die("Can't load tree %s", sha1_to_hex(sha1));
 957        }
 958
 959        c = buf;
 960        while (c != (buf + size)) {
 961                struct tree_entry *e = new_tree_entry();
 962
 963                if (t->entry_count == t->entry_capacity)
 964                        root->tree = t = grow_tree_content(t, 8);
 965                t->entries[t->entry_count++] = e;
 966
 967                e->tree = NULL;
 968                c = get_mode(c, &e->versions[1].mode);
 969                if (!c)
 970                        die("Corrupt mode in %s", sha1_to_hex(sha1));
 971                e->versions[0].mode = e->versions[1].mode;
 972                e->name = to_atom(c, strlen(c));
 973                c += e->name->str_len + 1;
 974                hashcpy(e->versions[0].sha1, (unsigned char*)c);
 975                hashcpy(e->versions[1].sha1, (unsigned char*)c);
 976                c += 20;
 977        }
 978        free(buf);
 979}
 980
 981static int tecmp0 (const void *_a, const void *_b)
 982{
 983        struct tree_entry *a = *((struct tree_entry**)_a);
 984        struct tree_entry *b = *((struct tree_entry**)_b);
 985        return base_name_compare(
 986                a->name->str_dat, a->name->str_len, a->versions[0].mode,
 987                b->name->str_dat, b->name->str_len, b->versions[0].mode);
 988}
 989
 990static int tecmp1 (const void *_a, const void *_b)
 991{
 992        struct tree_entry *a = *((struct tree_entry**)_a);
 993        struct tree_entry *b = *((struct tree_entry**)_b);
 994        return base_name_compare(
 995                a->name->str_dat, a->name->str_len, a->versions[1].mode,
 996                b->name->str_dat, b->name->str_len, b->versions[1].mode);
 997}
 998
 999static void mktree(struct tree_content *t,
1000        int v,
1001        unsigned long *szp,
1002        struct dbuf *b)
1003{
1004        size_t maxlen = 0;
1005        unsigned int i;
1006        char *c;
1007
1008        if (!v)
1009                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1010        else
1011                qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1012
1013        for (i = 0; i < t->entry_count; i++) {
1014                if (t->entries[i]->versions[v].mode)
1015                        maxlen += t->entries[i]->name->str_len + 34;
1016        }
1017
1018        size_dbuf(b, maxlen);
1019        c = b->buffer;
1020        for (i = 0; i < t->entry_count; i++) {
1021                struct tree_entry *e = t->entries[i];
1022                if (!e->versions[v].mode)
1023                        continue;
1024                c += sprintf(c, "%o", e->versions[v].mode);
1025                *c++ = ' ';
1026                strcpy(c, e->name->str_dat);
1027                c += e->name->str_len + 1;
1028                hashcpy((unsigned char*)c, e->versions[v].sha1);
1029                c += 20;
1030        }
1031        *szp = c - (char*)b->buffer;
1032}
1033
1034static void store_tree(struct tree_entry *root)
1035{
1036        struct tree_content *t = root->tree;
1037        unsigned int i, j, del;
1038        unsigned long new_len;
1039        struct last_object lo;
1040        struct object_entry *le;
1041
1042        if (!is_null_sha1(root->versions[1].sha1))
1043                return;
1044
1045        for (i = 0; i < t->entry_count; i++) {
1046                if (t->entries[i]->tree)
1047                        store_tree(t->entries[i]);
1048        }
1049
1050        le = find_object(root->versions[0].sha1);
1051        if (!S_ISDIR(root->versions[0].mode)
1052                || !le
1053                || le->pack_id != pack_id) {
1054                lo.data = NULL;
1055                lo.depth = 0;
1056        } else {
1057                mktree(t, 0, &lo.len, &old_tree);
1058                lo.data = old_tree.buffer;
1059                lo.offset = le->offset;
1060                lo.depth = t->delta_depth;
1061                lo.no_free = 1;
1062        }
1063
1064        mktree(t, 1, &new_len, &new_tree);
1065        store_object(OBJ_TREE, new_tree.buffer, new_len,
1066                &lo, root->versions[1].sha1, 0);
1067
1068        t->delta_depth = lo.depth;
1069        for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1070                struct tree_entry *e = t->entries[i];
1071                if (e->versions[1].mode) {
1072                        e->versions[0].mode = e->versions[1].mode;
1073                        hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1074                        t->entries[j++] = e;
1075                } else {
1076                        release_tree_entry(e);
1077                        del++;
1078                }
1079        }
1080        t->entry_count -= del;
1081}
1082
1083static int tree_content_set(
1084        struct tree_entry *root,
1085        const char *p,
1086        const unsigned char *sha1,
1087        const unsigned int mode)
1088{
1089        struct tree_content *t = root->tree;
1090        const char *slash1;
1091        unsigned int i, n;
1092        struct tree_entry *e;
1093
1094        slash1 = strchr(p, '/');
1095        if (slash1)
1096                n = slash1 - p;
1097        else
1098                n = strlen(p);
1099
1100        for (i = 0; i < t->entry_count; i++) {
1101                e = t->entries[i];
1102                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1103                        if (!slash1) {
1104                                if (e->versions[1].mode == mode
1105                                                && !hashcmp(e->versions[1].sha1, sha1))
1106                                        return 0;
1107                                e->versions[1].mode = mode;
1108                                hashcpy(e->versions[1].sha1, sha1);
1109                                if (e->tree) {
1110                                        release_tree_content_recursive(e->tree);
1111                                        e->tree = NULL;
1112                                }
1113                                hashclr(root->versions[1].sha1);
1114                                return 1;
1115                        }
1116                        if (!S_ISDIR(e->versions[1].mode)) {
1117                                e->tree = new_tree_content(8);
1118                                e->versions[1].mode = S_IFDIR;
1119                        }
1120                        if (!e->tree)
1121                                load_tree(e);
1122                        if (tree_content_set(e, slash1 + 1, sha1, mode)) {
1123                                hashclr(root->versions[1].sha1);
1124                                return 1;
1125                        }
1126                        return 0;
1127                }
1128        }
1129
1130        if (t->entry_count == t->entry_capacity)
1131                root->tree = t = grow_tree_content(t, 8);
1132        e = new_tree_entry();
1133        e->name = to_atom(p, n);
1134        e->versions[0].mode = 0;
1135        hashclr(e->versions[0].sha1);
1136        t->entries[t->entry_count++] = e;
1137        if (slash1) {
1138                e->tree = new_tree_content(8);
1139                e->versions[1].mode = S_IFDIR;
1140                tree_content_set(e, slash1 + 1, sha1, mode);
1141        } else {
1142                e->tree = NULL;
1143                e->versions[1].mode = mode;
1144                hashcpy(e->versions[1].sha1, sha1);
1145        }
1146        hashclr(root->versions[1].sha1);
1147        return 1;
1148}
1149
1150static int tree_content_remove(struct tree_entry *root, const char *p)
1151{
1152        struct tree_content *t = root->tree;
1153        const char *slash1;
1154        unsigned int i, n;
1155        struct tree_entry *e;
1156
1157        slash1 = strchr(p, '/');
1158        if (slash1)
1159                n = slash1 - p;
1160        else
1161                n = strlen(p);
1162
1163        for (i = 0; i < t->entry_count; i++) {
1164                e = t->entries[i];
1165                if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1166                        if (!slash1 || !S_ISDIR(e->versions[1].mode))
1167                                goto del_entry;
1168                        if (!e->tree)
1169                                load_tree(e);
1170                        if (tree_content_remove(e, slash1 + 1)) {
1171                                for (n = 0; n < e->tree->entry_count; n++) {
1172                                        if (e->tree->entries[n]->versions[1].mode) {
1173                                                hashclr(root->versions[1].sha1);
1174                                                return 1;
1175                                        }
1176                                }
1177                                goto del_entry;
1178                        }
1179                        return 0;
1180                }
1181        }
1182        return 0;
1183
1184del_entry:
1185        if (e->tree) {
1186                release_tree_content_recursive(e->tree);
1187                e->tree = NULL;
1188        }
1189        e->versions[1].mode = 0;
1190        hashclr(e->versions[1].sha1);
1191        hashclr(root->versions[1].sha1);
1192        return 1;
1193}
1194
1195static void dump_branches()
1196{
1197        static const char *msg = "fast-import";
1198        unsigned int i;
1199        struct branch *b;
1200        struct ref_lock *lock;
1201
1202        for (i = 0; i < branch_table_sz; i++) {
1203                for (b = branch_table[i]; b; b = b->table_next_branch) {
1204                        lock = lock_any_ref_for_update(b->name, NULL);
1205                        if (!lock || write_ref_sha1(lock, b->sha1, msg) < 0)
1206                                die("Can't write %s", b->name);
1207                }
1208        }
1209}
1210
1211static void dump_tags()
1212{
1213        static const char *msg = "fast-import";
1214        struct tag *t;
1215        struct ref_lock *lock;
1216        char path[PATH_MAX];
1217
1218        for (t = first_tag; t; t = t->next_tag) {
1219                sprintf(path, "refs/tags/%s", t->name);
1220                lock = lock_any_ref_for_update(path, NULL);
1221                if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1222                        die("Can't write %s", path);
1223        }
1224}
1225
1226static void dump_marks_helper(FILE *f,
1227        unsigned long base,
1228        struct mark_set *m)
1229{
1230        int k;
1231        if (m->shift) {
1232                for (k = 0; k < 1024; k++) {
1233                        if (m->data.sets[k])
1234                                dump_marks_helper(f, (base + k) << m->shift,
1235                                        m->data.sets[k]);
1236                }
1237        } else {
1238                for (k = 0; k < 1024; k++) {
1239                        if (m->data.marked[k])
1240                                fprintf(f, ":%lu %s\n", base + k,
1241                                        sha1_to_hex(m->data.marked[k]->sha1));
1242                }
1243        }
1244}
1245
1246static void dump_marks()
1247{
1248        if (mark_file)
1249        {
1250                FILE *f = fopen(mark_file, "w");
1251                dump_marks_helper(f, 0, marks);
1252                fclose(f);
1253        }
1254}
1255
1256static void read_next_command()
1257{
1258        read_line(&command_buf, stdin, '\n');
1259}
1260
1261static void cmd_mark()
1262{
1263        if (!strncmp("mark :", command_buf.buf, 6)) {
1264                next_mark = strtoul(command_buf.buf + 6, NULL, 10);
1265                read_next_command();
1266        }
1267        else
1268                next_mark = 0;
1269}
1270
1271static void* cmd_data (size_t *size)
1272{
1273        size_t n = 0;
1274        void *buffer;
1275        size_t length;
1276
1277        if (strncmp("data ", command_buf.buf, 5))
1278                die("Expected 'data n' command, found: %s", command_buf.buf);
1279
1280        length = strtoul(command_buf.buf + 5, NULL, 10);
1281        buffer = xmalloc(length);
1282
1283        while (n < length) {
1284                size_t s = fread((char*)buffer + n, 1, length - n, stdin);
1285                if (!s && feof(stdin))
1286                        die("EOF in data (%lu bytes remaining)", length - n);
1287                n += s;
1288        }
1289
1290        if (fgetc(stdin) != '\n')
1291                die("An lf did not trail the binary data as expected.");
1292
1293        *size = length;
1294        return buffer;
1295}
1296
1297static void cmd_new_blob()
1298{
1299        size_t l;
1300        void *d;
1301
1302        read_next_command();
1303        cmd_mark();
1304        d = cmd_data(&l);
1305
1306        if (store_object(OBJ_BLOB, d, l, &last_blob, NULL, next_mark))
1307                free(d);
1308}
1309
1310static void unload_one_branch()
1311{
1312        while (cur_active_branches
1313                && cur_active_branches >= max_active_branches) {
1314                unsigned long min_commit = ULONG_MAX;
1315                struct branch *e, *l = NULL, *p = NULL;
1316
1317                for (e = active_branches; e; e = e->active_next_branch) {
1318                        if (e->last_commit < min_commit) {
1319                                p = l;
1320                                min_commit = e->last_commit;
1321                        }
1322                        l = e;
1323                }
1324
1325                if (p) {
1326                        e = p->active_next_branch;
1327                        p->active_next_branch = e->active_next_branch;
1328                } else {
1329                        e = active_branches;
1330                        active_branches = e->active_next_branch;
1331                }
1332                e->active_next_branch = NULL;
1333                if (e->branch_tree.tree) {
1334                        release_tree_content_recursive(e->branch_tree.tree);
1335                        e->branch_tree.tree = NULL;
1336                }
1337                cur_active_branches--;
1338        }
1339}
1340
1341static void load_branch(struct branch *b)
1342{
1343        load_tree(&b->branch_tree);
1344        b->active_next_branch = active_branches;
1345        active_branches = b;
1346        cur_active_branches++;
1347        branch_load_count++;
1348}
1349
1350static void file_change_m(struct branch *b)
1351{
1352        const char *p = command_buf.buf + 2;
1353        char *p_uq;
1354        const char *endp;
1355        struct object_entry *oe;
1356        unsigned char sha1[20];
1357        unsigned int mode;
1358        char type[20];
1359
1360        p = get_mode(p, &mode);
1361        if (!p)
1362                die("Corrupt mode: %s", command_buf.buf);
1363        switch (mode) {
1364        case S_IFREG | 0644:
1365        case S_IFREG | 0755:
1366        case S_IFLNK:
1367        case 0644:
1368        case 0755:
1369                /* ok */
1370                break;
1371        default:
1372                die("Corrupt mode: %s", command_buf.buf);
1373        }
1374
1375        if (*p == ':') {
1376                char *x;
1377                oe = find_mark(strtoul(p + 1, &x, 10));
1378                hashcpy(sha1, oe->sha1);
1379                p = x;
1380        } else {
1381                if (get_sha1_hex(p, sha1))
1382                        die("Invalid SHA1: %s", command_buf.buf);
1383                oe = find_object(sha1);
1384                p += 40;
1385        }
1386        if (*p++ != ' ')
1387                die("Missing space after SHA1: %s", command_buf.buf);
1388
1389        p_uq = unquote_c_style(p, &endp);
1390        if (p_uq) {
1391                if (*endp)
1392                        die("Garbage after path in: %s", command_buf.buf);
1393                p = p_uq;
1394        }
1395
1396        if (oe) {
1397                if (oe->type != OBJ_BLOB)
1398                        die("Not a blob (actually a %s): %s",
1399                                command_buf.buf, type_names[oe->type]);
1400        } else {
1401                if (sha1_object_info(sha1, type, NULL))
1402                        die("Blob not found: %s", command_buf.buf);
1403                if (strcmp(blob_type, type))
1404                        die("Not a blob (actually a %s): %s",
1405                                command_buf.buf, type);
1406        }
1407
1408        tree_content_set(&b->branch_tree, p, sha1, S_IFREG | mode);
1409
1410        if (p_uq)
1411                free(p_uq);
1412}
1413
1414static void file_change_d(struct branch *b)
1415{
1416        const char *p = command_buf.buf + 2;
1417        char *p_uq;
1418        const char *endp;
1419
1420        p_uq = unquote_c_style(p, &endp);
1421        if (p_uq) {
1422                if (*endp)
1423                        die("Garbage after path in: %s", command_buf.buf);
1424                p = p_uq;
1425        }
1426        tree_content_remove(&b->branch_tree, p);
1427        if (p_uq)
1428                free(p_uq);
1429}
1430
1431static void cmd_from(struct branch *b)
1432{
1433        const char *from, *endp;
1434        char *str_uq;
1435        struct branch *s;
1436
1437        if (strncmp("from ", command_buf.buf, 5))
1438                return;
1439
1440        if (b->last_commit)
1441                die("Can't reinitailize branch %s", b->name);
1442
1443        from = strchr(command_buf.buf, ' ') + 1;
1444        str_uq = unquote_c_style(from, &endp);
1445        if (str_uq) {
1446                if (*endp)
1447                        die("Garbage after string in: %s", command_buf.buf);
1448                from = str_uq;
1449        }
1450
1451        s = lookup_branch(from);
1452        if (b == s)
1453                die("Can't create a branch from itself: %s", b->name);
1454        else if (s) {
1455                unsigned char *t = s->branch_tree.versions[1].sha1;
1456                hashcpy(b->sha1, s->sha1);
1457                hashcpy(b->branch_tree.versions[0].sha1, t);
1458                hashcpy(b->branch_tree.versions[1].sha1, t);
1459        } else if (*from == ':') {
1460                unsigned long idnum = strtoul(from + 1, NULL, 10);
1461                struct object_entry *oe = find_mark(idnum);
1462                unsigned long size;
1463                char *buf;
1464                if (oe->type != OBJ_COMMIT)
1465                        die("Mark :%lu not a commit", idnum);
1466                hashcpy(b->sha1, oe->sha1);
1467                buf = gfi_unpack_entry(oe, &size);
1468                if (!buf || size < 46)
1469                        die("Not a valid commit: %s", from);
1470                if (memcmp("tree ", buf, 5)
1471                        || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
1472                        die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1473                free(buf);
1474                hashcpy(b->branch_tree.versions[0].sha1,
1475                        b->branch_tree.versions[1].sha1);
1476        } else if (!get_sha1(from, b->sha1)) {
1477                if (is_null_sha1(b->sha1)) {
1478                        hashclr(b->branch_tree.versions[0].sha1);
1479                        hashclr(b->branch_tree.versions[1].sha1);
1480                } else {
1481                        unsigned long size;
1482                        char *buf;
1483
1484                        buf = read_object_with_reference(b->sha1,
1485                                type_names[OBJ_COMMIT], &size, b->sha1);
1486                        if (!buf || size < 46)
1487                                die("Not a valid commit: %s", from);
1488                        if (memcmp("tree ", buf, 5)
1489                                || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
1490                                die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1491                        free(buf);
1492                        hashcpy(b->branch_tree.versions[0].sha1,
1493                                b->branch_tree.versions[1].sha1);
1494                }
1495        } else
1496                die("Invalid ref name or SHA1 expression: %s", from);
1497
1498        read_next_command();
1499}
1500
1501static struct hash_list* cmd_merge(unsigned int *count)
1502{
1503        struct hash_list *list = NULL, *n, *e;
1504        const char *from, *endp;
1505        char *str_uq;
1506        struct branch *s;
1507
1508        *count = 0;
1509        while (!strncmp("merge ", command_buf.buf, 6)) {
1510                from = strchr(command_buf.buf, ' ') + 1;
1511                str_uq = unquote_c_style(from, &endp);
1512                if (str_uq) {
1513                        if (*endp)
1514                                die("Garbage after string in: %s", command_buf.buf);
1515                        from = str_uq;
1516                }
1517
1518                n = xmalloc(sizeof(*n));
1519                s = lookup_branch(from);
1520                if (s)
1521                        hashcpy(n->sha1, s->sha1);
1522                else if (*from == ':') {
1523                        unsigned long idnum = strtoul(from + 1, NULL, 10);
1524                        struct object_entry *oe = find_mark(idnum);
1525                        if (oe->type != OBJ_COMMIT)
1526                                die("Mark :%lu not a commit", idnum);
1527                        hashcpy(n->sha1, oe->sha1);
1528                } else if (get_sha1(from, n->sha1))
1529                        die("Invalid ref name or SHA1 expression: %s", from);
1530
1531                n->next = NULL;
1532                if (list)
1533                        e->next = n;
1534                else
1535                        list = n;
1536                e = n;
1537                *count++;
1538                read_next_command();
1539        }
1540        return list;
1541}
1542
1543static void cmd_new_commit()
1544{
1545        struct branch *b;
1546        void *msg;
1547        size_t msglen;
1548        char *str_uq;
1549        const char *endp;
1550        char *sp;
1551        char *author = NULL;
1552        char *committer = NULL;
1553        struct hash_list *merge_list = NULL;
1554        unsigned int merge_count;
1555
1556        /* Obtain the branch name from the rest of our command */
1557        sp = strchr(command_buf.buf, ' ') + 1;
1558        str_uq = unquote_c_style(sp, &endp);
1559        if (str_uq) {
1560                if (*endp)
1561                        die("Garbage after ref in: %s", command_buf.buf);
1562                sp = str_uq;
1563        }
1564        b = lookup_branch(sp);
1565        if (!b)
1566                b = new_branch(sp);
1567        if (str_uq)
1568                free(str_uq);
1569
1570        read_next_command();
1571        cmd_mark();
1572        if (!strncmp("author ", command_buf.buf, 7)) {
1573                author = strdup(command_buf.buf);
1574                read_next_command();
1575        }
1576        if (!strncmp("committer ", command_buf.buf, 10)) {
1577                committer = strdup(command_buf.buf);
1578                read_next_command();
1579        }
1580        if (!committer)
1581                die("Expected committer but didn't get one");
1582        msg = cmd_data(&msglen);
1583        read_next_command();
1584        cmd_from(b);
1585        merge_list = cmd_merge(&merge_count);
1586
1587        /* ensure the branch is active/loaded */
1588        if (!b->branch_tree.tree || !max_active_branches) {
1589                unload_one_branch();
1590                load_branch(b);
1591        }
1592
1593        /* file_change* */
1594        for (;;) {
1595                if (1 == command_buf.len)
1596                        break;
1597                else if (!strncmp("M ", command_buf.buf, 2))
1598                        file_change_m(b);
1599                else if (!strncmp("D ", command_buf.buf, 2))
1600                        file_change_d(b);
1601                else
1602                        die("Unsupported file_change: %s", command_buf.buf);
1603                read_next_command();
1604        }
1605
1606        /* build the tree and the commit */
1607        store_tree(&b->branch_tree);
1608        hashcpy(b->branch_tree.versions[0].sha1,
1609                b->branch_tree.versions[1].sha1);
1610        size_dbuf(&new_data, 97 + msglen
1611                + merge_count * 49
1612                + (author
1613                        ? strlen(author) + strlen(committer)
1614                        : 2 * strlen(committer)));
1615        sp = new_data.buffer;
1616        sp += sprintf(sp, "tree %s\n",
1617                sha1_to_hex(b->branch_tree.versions[1].sha1));
1618        if (!is_null_sha1(b->sha1))
1619                sp += sprintf(sp, "parent %s\n", sha1_to_hex(b->sha1));
1620        while (merge_list) {
1621                struct hash_list *next = merge_list->next;
1622                sp += sprintf(sp, "parent %s\n", sha1_to_hex(merge_list->sha1));
1623                free(merge_list);
1624                merge_list = next;
1625        }
1626        if (author)
1627                sp += sprintf(sp, "%s\n", author);
1628        else
1629                sp += sprintf(sp, "author %s\n", committer + 10);
1630        sp += sprintf(sp, "%s\n\n", committer);
1631        memcpy(sp, msg, msglen);
1632        sp += msglen;
1633        if (author)
1634                free(author);
1635        free(committer);
1636        free(msg);
1637
1638        store_object(OBJ_COMMIT,
1639                new_data.buffer, sp - (char*)new_data.buffer,
1640                NULL, b->sha1, next_mark);
1641        b->last_commit = object_count_by_type[OBJ_COMMIT];
1642
1643        if (branch_log) {
1644                int need_dq = quote_c_style(b->name, NULL, NULL, 0);
1645                fprintf(branch_log, "commit ");
1646                if (need_dq) {
1647                        fputc('"', branch_log);
1648                        quote_c_style(b->name, NULL, branch_log, 0);
1649                        fputc('"', branch_log);
1650                } else
1651                        fprintf(branch_log, "%s", b->name);
1652                fprintf(branch_log," :%lu %s\n",next_mark,sha1_to_hex(b->sha1));
1653        }
1654}
1655
1656static void cmd_new_tag()
1657{
1658        char *str_uq;
1659        const char *endp;
1660        char *sp;
1661        const char *from;
1662        char *tagger;
1663        struct branch *s;
1664        void *msg;
1665        size_t msglen;
1666        struct tag *t;
1667        unsigned long from_mark = 0;
1668        unsigned char sha1[20];
1669
1670        /* Obtain the new tag name from the rest of our command */
1671        sp = strchr(command_buf.buf, ' ') + 1;
1672        str_uq = unquote_c_style(sp, &endp);
1673        if (str_uq) {
1674                if (*endp)
1675                        die("Garbage after tag name in: %s", command_buf.buf);
1676                sp = str_uq;
1677        }
1678        t = pool_alloc(sizeof(struct tag));
1679        t->next_tag = NULL;
1680        t->name = pool_strdup(sp);
1681        if (last_tag)
1682                last_tag->next_tag = t;
1683        else
1684                first_tag = t;
1685        last_tag = t;
1686        if (str_uq)
1687                free(str_uq);
1688        read_next_command();
1689
1690        /* from ... */
1691        if (strncmp("from ", command_buf.buf, 5))
1692                die("Expected from command, got %s", command_buf.buf);
1693
1694        from = strchr(command_buf.buf, ' ') + 1;
1695        str_uq = unquote_c_style(from, &endp);
1696        if (str_uq) {
1697                if (*endp)
1698                        die("Garbage after string in: %s", command_buf.buf);
1699                from = str_uq;
1700        }
1701
1702        s = lookup_branch(from);
1703        if (s) {
1704                hashcpy(sha1, s->sha1);
1705        } else if (*from == ':') {
1706                from_mark = strtoul(from + 1, NULL, 10);
1707                struct object_entry *oe = find_mark(from_mark);
1708                if (oe->type != OBJ_COMMIT)
1709                        die("Mark :%lu not a commit", from_mark);
1710                hashcpy(sha1, oe->sha1);
1711        } else if (!get_sha1(from, sha1)) {
1712                unsigned long size;
1713                char *buf;
1714
1715                buf = read_object_with_reference(sha1,
1716                        type_names[OBJ_COMMIT], &size, sha1);
1717                if (!buf || size < 46)
1718                        die("Not a valid commit: %s", from);
1719                free(buf);
1720        } else
1721                die("Invalid ref name or SHA1 expression: %s", from);
1722
1723        if (str_uq)
1724                free(str_uq);
1725        read_next_command();
1726
1727        /* tagger ... */
1728        if (strncmp("tagger ", command_buf.buf, 7))
1729                die("Expected tagger command, got %s", command_buf.buf);
1730        tagger = strdup(command_buf.buf);
1731
1732        /* tag payload/message */
1733        read_next_command();
1734        msg = cmd_data(&msglen);
1735
1736        /* build the tag object */
1737        size_dbuf(&new_data, 67+strlen(t->name)+strlen(tagger)+msglen);
1738        sp = new_data.buffer;
1739        sp += sprintf(sp, "object %s\n", sha1_to_hex(sha1));
1740        sp += sprintf(sp, "type %s\n", type_names[OBJ_COMMIT]);
1741        sp += sprintf(sp, "tag %s\n", t->name);
1742        sp += sprintf(sp, "%s\n\n", tagger);
1743        memcpy(sp, msg, msglen);
1744        sp += msglen;
1745        free(tagger);
1746        free(msg);
1747
1748        store_object(OBJ_TAG, new_data.buffer, sp - (char*)new_data.buffer,
1749                NULL, t->sha1, 0);
1750
1751        if (branch_log) {
1752                int need_dq = quote_c_style(t->name, NULL, NULL, 0);
1753                fprintf(branch_log, "tag ");
1754                if (need_dq) {
1755                        fputc('"', branch_log);
1756                        quote_c_style(t->name, NULL, branch_log, 0);
1757                        fputc('"', branch_log);
1758                } else
1759                        fprintf(branch_log, "%s", t->name);
1760                fprintf(branch_log," :%lu %s\n",from_mark,sha1_to_hex(t->sha1));
1761        }
1762}
1763
1764static void cmd_reset_branch()
1765{
1766        struct branch *b;
1767        char *str_uq;
1768        const char *endp;
1769        char *sp;
1770
1771        /* Obtain the branch name from the rest of our command */
1772        sp = strchr(command_buf.buf, ' ') + 1;
1773        str_uq = unquote_c_style(sp, &endp);
1774        if (str_uq) {
1775                if (*endp)
1776                        die("Garbage after ref in: %s", command_buf.buf);
1777                sp = str_uq;
1778        }
1779        b = lookup_branch(sp);
1780        if (b) {
1781                b->last_commit = 0;
1782                if (b->branch_tree.tree) {
1783                        release_tree_content_recursive(b->branch_tree.tree);
1784                        b->branch_tree.tree = NULL;
1785                }
1786        }
1787        else
1788                b = new_branch(sp);
1789        if (str_uq)
1790                free(str_uq);
1791        read_next_command();
1792        cmd_from(b);
1793}
1794
1795static void cmd_checkpoint()
1796{
1797        if (object_count)
1798                checkpoint();
1799        read_next_command();
1800}
1801
1802static const char fast_import_usage[] =
1803"git-fast-import [--objects=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file] [--branch-log=log] temp.pack";
1804
1805int main(int argc, const char **argv)
1806{
1807        int i;
1808        unsigned long est_obj_cnt = object_entry_alloc;
1809        unsigned long duplicate_count;
1810
1811        setup_ident();
1812        git_config(git_default_config);
1813
1814        for (i = 1; i < argc; i++) {
1815                const char *a = argv[i];
1816
1817                if (*a != '-' || !strcmp(a, "--"))
1818                        break;
1819                else if (!strncmp(a, "--objects=", 10))
1820                        est_obj_cnt = strtoul(a + 10, NULL, 0);
1821                else if (!strncmp(a, "--max-objects-per-pack=", 23))
1822                        max_objects = strtoul(a + 23, NULL, 0);
1823                else if (!strncmp(a, "--max-pack-size=", 16))
1824                        max_packsize = strtoul(a + 16, NULL, 0) * 1024 * 1024;
1825                else if (!strncmp(a, "--depth=", 8))
1826                        max_depth = strtoul(a + 8, NULL, 0);
1827                else if (!strncmp(a, "--active-branches=", 18))
1828                        max_active_branches = strtoul(a + 18, NULL, 0);
1829                else if (!strncmp(a, "--export-marks=", 15))
1830                        mark_file = a + 15;
1831                else if (!strncmp(a, "--branch-log=", 13)) {
1832                        branch_log = fopen(a + 13, "w");
1833                        if (!branch_log)
1834                                die("Can't create %s: %s", a + 13, strerror(errno));
1835                }
1836                else
1837                        die("unknown option %s", a);
1838        }
1839        if ((i+1) != argc)
1840                usage(fast_import_usage);
1841        base_name = argv[i];
1842
1843        alloc_objects(est_obj_cnt);
1844        strbuf_init(&command_buf);
1845
1846        atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
1847        branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
1848        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
1849        marks = pool_calloc(1, sizeof(struct mark_set));
1850
1851        start_packfile();
1852        for (;;) {
1853                read_next_command();
1854                if (command_buf.eof)
1855                        break;
1856                else if (!strcmp("blob", command_buf.buf))
1857                        cmd_new_blob();
1858                else if (!strncmp("commit ", command_buf.buf, 7))
1859                        cmd_new_commit();
1860                else if (!strncmp("tag ", command_buf.buf, 4))
1861                        cmd_new_tag();
1862                else if (!strncmp("reset ", command_buf.buf, 6))
1863                        cmd_reset_branch();
1864                else if (!strcmp("checkpoint", command_buf.buf))
1865                        cmd_checkpoint();
1866                else
1867                        die("Unsupported command: %s", command_buf.buf);
1868        }
1869        end_packfile();
1870
1871        dump_branches();
1872        dump_tags();
1873        dump_marks();
1874        if (branch_log)
1875                fclose(branch_log);
1876
1877        for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
1878                duplicate_count += duplicate_count_by_type[i];
1879
1880        fprintf(stderr, "%s statistics:\n", argv[0]);
1881        fprintf(stderr, "---------------------------------------------------------------------\n");
1882        fprintf(stderr, "Alloc'd objects: %10lu (%10lu overflow  )\n", alloc_count, alloc_count - est_obj_cnt);
1883        fprintf(stderr, "Total objects:   %10lu (%10lu duplicates                  )\n", object_count, duplicate_count);
1884        fprintf(stderr, "      blobs  :   %10lu (%10lu duplicates %10lu deltas)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB]);
1885        fprintf(stderr, "      trees  :   %10lu (%10lu duplicates %10lu deltas)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE]);
1886        fprintf(stderr, "      commits:   %10lu (%10lu duplicates %10lu deltas)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT]);
1887        fprintf(stderr, "      tags   :   %10lu (%10lu duplicates %10lu deltas)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG]);
1888        fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
1889        fprintf(stderr, "      marks:     %10u (%10lu unique    )\n", (1 << marks->shift) * 1024, marks_set_count);
1890        fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
1891        fprintf(stderr, "Memory total:    %10lu KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
1892        fprintf(stderr, "       pools:    %10lu KiB\n", total_allocd/1024);
1893        fprintf(stderr, "     objects:    %10lu KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
1894        fprintf(stderr, "---------------------------------------------------------------------\n");
1895        pack_report();
1896        fprintf(stderr, "---------------------------------------------------------------------\n");
1897        fprintf(stderr, "\n");
1898
1899        return 0;
1900}